PbLite Contour Detection

---

Overview

Boundary or contour detection is an important problem in computer vision and is well studied. Other important problems like hierarchical image segmentation can be reduced to a contour detection problem. A version of the state-of-the-art edge detection algorithm Pb-Lite is discussed in this project. The Pb Lite approach causes a jump in the performance of boundary detection by suppressing false positives in the textured regions due to the coupling of multiscale local texture, brightness, and color information with a globalization framework.

Dependencies

• OpenCV: An Open Source Computer Vision Library released under a BSD license. A complete installation guide for OpenCV can be found here.

• numpy

• scipy

• matplotlib

• sklearn

• imutils

Standard install via command-line

git clone --recursive https://github.com/rohithjayarajan/PbLite-Contour-Detection.git

Run program for 10 images in the BSDS Images folder:

cd <path to repository>
python ./Code/Wrapper.py

A detailed report can be found in Report.pdf